MySQL 5.5.8 in production – First impressions

As I recently noted, we upgraded our Wix.com production databases to the newly released MySQL 5.5.8 and moved it to Solaris 10.

Here is the first experience and notes:

1. “mysqldump –single-transaction –flush-logs” does not produce consistent dump anymore
   This way I usually set up a new first slave. After I restored first three databases, I tried to set up replication using “–master-data=2” output from the dump file. Starting replication on the newly created slave has immediately failed with error 1062 – “show slave status\G” says: "1062 | Error 'Duplicate entry '...' for key 1' on query. Default database: '...'. Query: 'INSERT INTO ...". This never happened in previous versions. I don’t know yet exactly what happens there. Brief checking of a few tables is showing that all records are in place. Perhaps, several transactions were flushed twice and small portion of the transaction log was replayed on existing data. I carefully checked new mysqldump documentation, there is no change in –single-transaction behavior. I made a post in the MySQL forum. The forum is somewhat sleepy, waiting for replies…
   Reading mysqldump documentation has revealed a few new quite useful features added in mysqldump in version 5.5: 
   • --dump-slave[=value] – This option is very useful for cloning existing slave replicating data from the same master. It causes mysqldump to include “change master” statement exactly how it does –master-data, but the master info is of the dumped slave’s master. If the running MySQL slave cannotbe stopped for faster cloning procedure, using this option can be very handy.
   • –apply-slave-statements – This one makes life easier in conjunction with –dump-slave when resulting dump should be reapplied on existing slave. It says to mysqldump to include STOP SLAVE prior to CHANGE MASTER statement and START SLAVE at end of the resulting dump.
2. InnoDB Table Compression works even better than expected
   Compression seems to work quite smooth and stable. Using 16K pages for table containing large XML blobs helped to reduce 1TB database to approximately 250G which was expected. But performance gain is pretty surprising. Look at the New Relic chart:
   
   Note the brown graph. This is behavior of most popular get XML call. We switched app server to the new compressed database at 3:00. Since then fetching blobs from the database is much faster.
3. Solaris packages still do not have DTrace probes enabled
   This is bad surprise. One of my big expectations after Sun has acquired MySQL is that Solaris support in MySQL will be significantly improved. I reported about lack of DTrace in 5.5.7-RC in the past. I was really hoping that this is just a matter of release candidate build and that will be solved when 5.5 will go to the GA line. After 5.5.8 was deployed on our servers we have found that there is is still empty “dtrace -l | grep mysql”. I made another post about it in the MySQL forum. No replies so far.
   So while “What Is New in MySQL 5.5” article is announcing “Enhanced Solaris Support“, in practice it is not that enhanced. There is also a good news regarding Solaris support. “innodb_flush_method = O_DIRECT” is now working (it was causing MySQL to produce an error on startup in 5.5.7-RC).

The bottom line so far is that MySQL 5.58 is ready for production. This is a big jump for the MySQL community. We can say good bye to the buggy 5.1 line and really enjoy the next generation of this amazing piece of of software.